Blood transfusion pipeline jitter detection method based on intelligent sensor

Abstract

The invention relates to the technical field of blood transfusion pipeline detection, and discloses a blood transfusion pipeline jitter detection method based on an intelligent sensor. According to the method, an initial control instruction generation layer receives an operation instruction, and a natural language analysis unit identifies a pipeline number and a preset jitter threshold parameter; the multi-mode signal detection layer collects pressure waveform signals and three-dimensional motion track signals of a blood transfusion pipeline in real time, and establishes a feature matching database to store reference signal feature combinations corresponding to different pipeline models. The dynamic mapping layer combines the real-time detection signal features with the reference signal features to perform similarity matching; the double-domain interpolation calculation layer generates an anti-jitter control parameter according to a matching result; the anti-phase jitter elimination execution layer outputs a phase compensation instruction according to the anti-jitter control parameters; the parameter optimization layer updates a feature matching database according to the historical jitter residual volume data; and the abnormal decision-making layer judges an unrestrained jitter state and triggers an alarm protocol, and also implements a full-period updating mechanism to refresh a reference signal feature combination.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of blood transfusion pipeline detection, in particular to a blood transfusion pipeline shaking detection method based on an intelligent sensor. BACKGROUND

[0002] In the modern medical system, the blood transfusion pipeline system is the key carrier of the core treatment links such as clinical blood transfusion, intensive care, and drug infusion, and its running stability directly determines the safety and effectiveness of treatment. In actual application, the blood transfusion pipeline is easily affected by external environmental interference (such as medical staff operation collision, patient body position change, and infusion pump mechanical vibration) and internal fluid characteristics (such as blood viscosity change and flow rate fluctuation) to produce shaking phenomenon. This shaking not only destroys the steady flow of the fluid in the pipeline, leading to infusion rate deviation, for example, in precise chemotherapy drug infusion, 0.5ml / min rate fluctuation may cause drug concentration to exceed the treatment window, increasing the risk of side effects; more likely to cause pipeline interface loosening to cause liquid leakage, or cause red blood cell rupture due to severe shaking to cause hemolysis reaction, which is serious and endangers the life of the patient. Therefore, real-time and accurate detection and inhibition of the shaking state of the blood transfusion pipeline have become a core technical problem to be solved in the field of medical equipment. The technical solutions for blood transfusion pipeline shaking detection in the industry mainly fall into three categories: the first is a single pressure signal detection method, which sets a piezoelectric pressure sensor at the pipeline joint to monitor the instantaneous fluctuation of the fluid pressure in the pipeline, and determines the shaking state when the pressure value exceeds the preset threshold. The core principle of this method is based on the physical characteristics that shaking will cause fluid turbulence, which in turn causes pressure mutation, but its limitations are significant. The pressure signal is easily disturbed by fluid viscosity, temperature, and pipeline material elastic coefficient, for example, when infusing high-viscosity blood products, normal fluid resistance changes may be misjudged as shaking; at the same time, for pipeline transverse displacement type shaking (such as horizontal swinging of the pipeline due to external force pulling), its influence on the pressure in the pipeline is minimal, and it is easy to miss detection. According to clinical data statistics, the combined false positive rate and false negative rate of this method can reach more than 23%. This technology captures the displacement trajectory of the outer wall of the pipeline by setting up a high-definition camera and image recognition algorithm next to the pipeline, and judges the shaking by calculating whether the displacement exceeds the set range. Its advantage is that it can directly monitor the physical movement of the pipeline, but in practical application it faces multiple technical bottlenecks: on the one hand, changes in environmental lighting (such as the adjustment of the angle of the operating room shadowless lamp, the fluctuation of natural light intensity in the ward) will cause distortion of the image gray value, making the trajectory recognition accuracy decrease; on the other hand, this method can only monitor the macro movement of the pipeline surface and cannot perceive the micro vibration caused by the fluid pulsation inside the pipeline (such as the high-frequency micro vibration of the pipeline caused by the movement of the infusion pump piston), and such micro vibration may cause blood component sedimentation and affect the quality of blood transfusion. In addition, the optical device needs to maintain a fixed distance from the pipeline, which is difficult to adapt to the diversified pipeline layout scenes in the clinic (such as bed-mounted pipeline, mobile infusion stand pipeline), and the applicability is limited. Some technical solutions try to combine pressure and optical signals, but still cannot break through the core technical bottlenecks: first, the reference signal characteristics are fixed, and existing systems usually only preset reference characteristics for a single type of pipeline (such as standard 12mm inner diameter infusion tube), while in clinical practice, there are various types of blood transfusion pipelines (such as 8mm pediatric special pipeline, 16mm large flow blood transfusion pipeline), and the differences in material stiffness and inner diameter size of different pipelines will cause significant differences in pressure waveform and movement trajectory when they are working normally. Fixed reference characteristics will increase the similarity matching error and reduce the detection accuracy; second, the anti-shake control parameters are static, and the control parameters (such as phase compensation amplitude) of existing systems are usually preset at the factory, and cannot be dynamically adjusted according to the real-time shaking intensity and type (such as low-frequency large-amplitude shaking, high-frequency small-amplitude shaking), which leads to unstable shaking elimination effect. For example, in the face of sudden severe pipeline pulling, fixed parameters may not be able to quickly suppress shaking, while in the face of slight fluid pulsation, excessive compensation may affect normal infusion; third, there is a lack of adaptive optimization mechanism, and the system cannot update the reference characteristics using historical shaking data (such as shaking residual amount in different patients and different treatment scenarios), which may lead to detection accuracy decay due to pipeline aging, sensor drift, etc. after long-term use; fourth, the abnormal decision logic is single, and existing alarm systems are mostly based on the judgment standard of "single shaking value exceeding limit", without considering the duration and frequency of shaking, which may cause problems such as "frequent alarm for slight shaking" or "missed alarm for intermittent strong shaking".

[0003] The prior art generally lacks a full-cycle reference update mechanism. When the pipeline model is replaced or the use environment changes, the reference parameters need to be manually recalibrated, which is tedious and prone to human error. At the same time, for uncontrollable extreme shaking (such as violent vibration caused by pipeline rupture), the existing system can only trigger a simple sound-light alarm and cannot link to the subsequent emergency handling process (such as automatically pausing the infusion pump and notifying medical staff), resulting in a lag in emergency response. In summary, the current infusion pipeline shaking detection technology has obvious deficiencies in detection comprehensiveness, precision adaptability, control dynamics, and decision intelligence, and cannot meet the clinical demand for high-precision, high-reliability, and self-adaptive shaking detection and suppression. A new detection method that can integrate multi-modal signals, dynamically update reference features, and self-adaptively adjust control parameters is needed to improve the operational safety and treatment security of the infusion pipeline system. SUMMARY

[0004] The purpose of the present application is to provide an intelligent sensor-based infusion pipeline shaking detection method to solve the problems raised in the background art.

[0005] To achieve the above-mentioned purpose, the present application provides an intelligent sensor-based infusion pipeline shaking detection method, which comprises: An initial control instruction generation layer is set to receive operation instructions, and a natural language analysis unit is used to identify the pipeline number and the preset shaking threshold parameter; A multi-modal signal detection layer is set to real-time collect the pressure waveform signal and three-dimensional motion trajectory signal of the infusion pipeline; a feature matching database is established to store the reference signal feature combination corresponding to different pipeline models; A dynamic mapping layer is set to perform similarity matching between the real-time detection signal features and the reference signal feature combination; A dual-domain interpolation calculation layer is set to generate anti-shaking control parameters according to the similarity matching result; An anti-phase anti-shaking execution layer is set to output phase compensation instructions according to the anti-shaking control parameters; A parameter optimization layer is set to update the feature matching database according to historical shaking residual data; An abnormality decision layer is set to judge the uncontrollable shaking state and trigger the alarm protocol, and a full-cycle update mechanism is implemented to refresh the reference signal feature combination according to the optimization command.

[0006] Preferably, the implementation process of the multi-modal signal detection layer includes: acquiring the frequency spectrum distribution feature of the pressure waveform signal through an array pressure sensor; simultaneously capturing the time-domain mutation feature of the three-dimensional motion trajectory signal using an inertial measurement unit; the frequency spectrum distribution feature and the time-domain mutation feature form a real-time detection signal feature set; the real-time detection signal feature set is transmitted to the dynamic mapping layer for similarity matching.

[0007] Preferably, the feature matching database construction process comprises: establishing a three-dimensional parameter framework for each pipeline number, the first dimension storing the reference pressure spectrum feature vector, the second dimension storing the reference motion trajectory feature vector, and the third dimension storing the optimal anti-jitter parameter combination; the reference pressure spectrum feature vector contains fundamental frequency and harmonic energy distribution data, and the reference motion trajectory feature vector contains displacement mutation amplitude and time sequence correlation data.

[0008] Preferably, the dynamic mapping layer execution process comprises: predefining a dynamic time warping path as a set of similarity anchor points; calculating the dynamic time warping distance of the real-time detection signal feature and the reference signal feature combination; when the dynamic time warping distance is less than a preset similarity threshold, outputting the matching successful reference signal feature combination index number; when the preset similarity threshold is exceeded, starting a nearest neighbor projection mechanism to select the reference signal feature combination with the smallest distance.

[0009] Preferably, the dual-domain interpolation calculation layer running mechanism comprises: in the feature domain interpolation stage, based on the matching successful reference signal feature combination index number, weighting and fusing a plurality of reference pressure spectrum feature vectors to generate a hybrid spectrum feature; in the device domain interpolation stage, performing coordinate transformation on the reference motion trajectory feature vector according to the installation inclination angle parameter of the current blood pipeline; finally, fusing the hybrid spectrum feature and the coordinate transformation result to generate the anti-jitter control parameter.

[0010] Preferably, the control process of the anti-phase jitter elimination execution layer comprises: receiving the anti-jitter control parameter generated by the dual-domain interpolation calculation layer; generating an anti-phase waveform instruction according to the phase offset parameter in the anti-jitter control parameter; adjusting the output gain of the piezoelectric ceramic actuator according to the amplitude suppression coefficient; the phase offset parameter and the amplitude suppression coefficient jointly constitute a phase compensation instruction.

[0011] Preferably, the method further comprises: setting a closed-loop feedback layer to monitor the residual amount of the compensated pipeline jitter in real time; The implementation of the closed-loop feedback layer comprises: after the anti-phase jitter elimination execution layer outputs the phase compensation instruction, reacquiring the pressure waveform signal to calculate the residual jitter energy value; comparing the residual jitter energy value with a preset jitter threshold parameter in real time; when the residual jitter energy value is continuously higher than the preset jitter threshold parameter, generating a feature matching database update request.

[0012] Preferably, the method further comprises: setting a parameter optimization layer to update the feature matching database according to historical jitter residual amount data; The update mechanism of the parameter optimization layer comprises: receiving a residual jitter energy value time sequence generated by the closed-loop feedback layer; constructing a gradient boosting decision tree model to analyze the association rules of the residual jitter energy value and the reference signal feature combination; when a specific reference pressure spectrum feature vector is detected to be related to high-frequency residual jitter, a feature matching database optimization command is generated; and the feature matching database optimization command triggers the full-cycle update mechanism.

[0013] Preferably, the judgment logic of the abnormal decision layer comprises: when residual jitter energy values in three consecutive detection cycles are all more than a preset percentage of a preset jitter threshold parameter, determining that it is an uncontrolled jitter state; generating a pipeline safety risk level assessment report; triggering an audible-light alarm protocol and sending an emergency shutdown instruction to the peristaltic pump control system.

[0014] Preferably, the execution steps of the full-cycle update mechanism comprise: receiving a feature matching database optimization command generated by the parameter optimization layer; extracting a real-time detection signal feature set corresponding to a time period of the latest residual jitter energy value; writing the real-time detection signal feature set as a new reference signal feature combination into the feature matching database; and resetting the coverage range of the similarity anchor point set of the dynamic mapping layer.

[0015] Compared with the prior art, the present application has the following beneficial effects: The blood transfusion pipeline jitter detection method based on an intelligent sensor realizes comprehensive optimization of blood transfusion pipeline jitter detection and anti-jitter control through multi-level cooperation. The initial control instruction generation layer is provided with a natural language analysis unit, which can directly identify the pipeline number and preset jitter threshold parameter in the operation instruction, without the need for manual input of complex parameters, thereby simplifying the operation process, reducing the operation difficulty of the staff, enabling personnel with different operation experiences to quickly complete parameter setting, and improving the ease of use of the detection system.

[0016] Compared with the traditional single signal acquisition mode, the combination of the two signals can more comprehensively reflect the pipeline operation state, the pressure waveform signal can capture the pressure change in the liquid transmission process, the three-dimensional motion trajectory signal can intuitively reflect the spatial position fluctuation of the pipeline, and the two signals complement each other to effectively distinguish signal changes caused by jitter and other factors, thereby reducing the occurrence of misjudgment. The establishment of the feature matching database provides corresponding reference signal feature combinations for different pipeline models, so that the detection system can accurately call the matched reference features according to the pipeline number, solves the poor adaptability of the existing detection method to different pipeline models, and improves the stability of the detection accuracy in different pipeline scenes. The dynamic mapping layer combines the real-time detection signal features with the reference signal features for similarity matching, can quickly judge whether the current pipeline signal deviates from the normal range, provides accurate judgment basis for subsequent anti-jitter control, and avoids the anti-jitter lag caused by untimely signal analysis. The dual-domain interpolation calculation layer generates anti-jitter control parameters according to the similarity matching result, and does not use fixed parameters for adjustment, but dynamically generates control parameters suitable for the current jitter situation in combination with the real-time signal difference, so that the anti-jitter control is more targeted, the anti-jitter effect is effectively improved, and the influence of jitter on liquid transmission is reduced. The anti-phase jitter elimination execution layer outputs a phase compensation instruction according to the anti-jitter control parameter, actively offsets the pipeline jitter through phase compensation, compared with passive adjustment, the active compensation can more quickly and directly suppress the jitter, and further guarantees the stability of the pipeline operation. The parameter optimization layer updates the feature matching database according to the historical jitter residual data, so that the reference signal feature combination can be continuously optimized with the accumulation of actual operation data, and gradually adapt to the pipeline operation characteristics in the current use environment. Even if the pipeline is slightly aged or the environmental conditions change after long-term use, the detection system can still maintain high detection accuracy and anti-jitter effect. The abnormality decision layer can timely judge the uncontainable jitter state and trigger the alarm protocol, so that the staff can quickly know the abnormal situation and take corresponding measures to avoid more serious consequences caused by uncontainable jitter; at the same time, the full-cycle updating mechanism is implemented to refresh the reference signal feature combination according to the optimization command, so as to ensure that the reference features are always in the optimal state, further prolong the effective service life of the detection system, improve its reliability and stability in the long-term use process, and better meet the demand for stable operation of the blood transfusion pipeline in medical and other scenes. BRIEF DESCRIPTION OF DRAWINGS

[0017] Figure 1 The working principle diagram of the blood transfusion pipeline jitter detection method based on the intelligent sensor is described. Figure 2 The working flowchart of the multi-modal signal detection layer implementation process is described. Figure 3 The working flowchart of the dynamic mapping layer implementation process is described. Figure 4 The working flowchart of the anti-phase jitter elimination execution layer control process is described. DETAILED DESCRIPTION

[0018] The technical solutions in the embodiments of the present application will be described clearly and completely below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor are within the scope of protection of the present application.

[0019] Referring to Figure 1 The application provides a smart sensor-based transfusion pipeline jitter detection method, which realizes real-time detection, matching, compensation and optimization of jitter through a multi-layer structure. The specific embodiments of the application are described in detail below.

[0020] An initial control instruction generation layer is arranged to receive operation instructions, and a natural language analysis unit is arranged to identify pipeline numbers and preset jitter threshold parameters. A multi-modal signal detection layer is arranged to collect pressure waveform signals and three-dimensional motion trajectory signals of the transfusion pipeline in real time. A feature matching database is established to store reference signal feature combinations corresponding to different pipeline models. A dynamic mapping layer is arranged to perform similarity matching between real-time detection signal features and reference signal feature combinations. A dual-domain interpolation calculation layer is arranged to generate anti-jitter control parameters according to the similarity matching results. An anti-phase jitter elimination execution layer is arranged to output phase compensation instructions according to the anti-jitter control parameters. A parameter optimization layer is arranged to update the feature matching database according to historical jitter residual data. An abnormality decision layer is arranged to determine an uncontainable jitter state and trigger an alarm protocol, and a full-cycle updating mechanism is arranged to refresh the reference signal feature combinations according to the optimization commands.

[0021] Embodiment 1: Referring to Figure 2 The implementation of the multi-modal signal detection layer is realized by hardware and software cooperation. Array type pressure sensors are arranged at specific monitoring points on the outer wall of the transfusion pipeline in an equal interval distribution manner, and each monitoring point corresponds to an independent signal acquisition channel. The capture of pressure waveform signals uses a differential amplification circuit to eliminate environmental noise interference, and the original analog signal is sampled by a 24-bit high-precision analog-to-digital converter to generate discrete time sequence data. The extraction process of the frequency spectrum distribution feature applies a fast Fourier transform algorithm to convert the time domain pressure data into frequency domain energy distribution. The process calculates the amplitude of the fundamental frequency component, the energy proportion of the second to fifth harmonic, and the frequency band energy cumulative value. The fundamental frequency component corresponds to the main frequency of the liquid pulsation in the pipeline, and the harmonic energy distribution reflects the nonlinear vibration characteristics. The frequency spectrum feature vector is updated according to a fixed time window, and the window length is set to 500 milliseconds with an overlap rate of 50%, ensuring real-time performance.

[0022] The inertial measurement unit running synchronously contains a three-axis accelerometer and a three-axis gyroscope to collect the spatial motion data of the blood transfusion pipeline at a sampling rate of 200 Hz. The pre-processing of the three-dimensional motion trajectory signal adopts the Kalman filter algorithm to fuse acceleration and angular velocity data to eliminate sensor drift errors. The identification of time-domain mutation features is based on the first-order derivative analysis of the motion trajectory: the displacement change rate of adjacent sampling points is calculated, and when the change rate exceeds the preset mutation threshold, the displacement mutation amplitude and duration at that time point are recorded. The time series correlation data is calculated by the autocorrelation function, and the similarity of the displacement sequence under a fixed time offset is analyzed. The final generated real-time detection signal feature set contains two types of data structures: pressure spectrum feature vectors (fundamental frequency amplitude, harmonic energy distribution array) and motion trajectory feature vectors (displacement mutation amplitude list, time series autocorrelation coefficient). The set is transmitted to the dynamic mapping layer through a high-speed serial communication interface.

[0023] The construction of the feature matching database adopts a hierarchical storage architecture. For each pipeline number, an independent storage partition is allocated, and a three-dimensional parameter matrix is established within the partition: the first dimension stores the reference pressure spectrum feature vector, and the vector dimension is consistent with the real-time detection features. The reference data comes from the laboratory calibration process: under standard temperature and flow rate conditions, the steady-state pressure signals of different pipeline models are collected, and the typical values and allowable fluctuation ranges of the fundamental frequency and harmonic energy distribution are generated through statistical analysis. The second dimension stores the reference motion trajectory feature vector, which includes the displacement mutation amplitude threshold matrix and the time series correlation template. The displacement mutation amplitude threshold is obtained by statistical analysis of historical data to obtain the maximum value and standard deviation, and the time series correlation template is discretized from the autocorrelation function curve of the typical motion trajectory. The third dimension stores the optimal anti-jitter parameter combination, including the phase compensation angle parameter (0°-360°) and the amplitude suppression gain coefficient (0.0-1.0). This parameter combination is determined through pre-test on the vibration table, and different parameter combinations are adjusted until the residual jitter energy is minimized during the test.

[0024] The database is implemented using a relational data table structure, with the pipeline number string as the primary key. Each number corresponds to multiple records, and each record contains a timestamped reference signal feature combination version. The data access interface provides fast retrieval based on index numbers, supporting parallel queries in real-time matching processes. Data is stored persistently in non-volatile memory, and update operations are guaranteed to be consistent through transaction mechanisms. During the initialization loading phase, the database preloads reference data for 20 standard pipeline models, with each model containing 10 feature combinations under different working conditions. During operation, new reference signal feature combinations are written in an appended manner, and the version number is automatically incremented.

[0025] The deployment of the array pressure sensor follows specific topology rules: three groups of sensor arrays are set at the inlet, mid-curved section and outlet of the pipeline, each group contains 4 sensing units arranged in a 90° ring. The sensing units are piezoresistive MEMS chips with a range of -5kPa to +5kPa and a linearity error less than 0.5% FS. The signal conditioning circuit contains a temperature compensation module to eliminate the zero drift caused by ambient temperature changes. The inertial measurement unit is installed on a rigid support frame in the middle section of the pipeline, with a distance of 15 cm from the pressure sensor array. The motion data acquisition uses a digital I 2 C interface, and the raw data is uniformly converted to the global coordinate system with the pipeline inlet as the origin through coordinate transformation.

[0026] The generation process of the real-time detection signal feature set includes a data verification step: after generating the pressure spectrum feature vector, verify whether the fundamental frequency component is within the physiological allowable frequency range (1-10 Hz); after calculating the motion trajectory feature vector, verify that the displacement mutation amplitude does not exceed the elastic deformation limit of the pipeline material. If the verification fails, the sensor self-diagnosis program will be triggered, and after excluding hardware faults, the data will be reacquired. The feature set transmission uses a fixed-length data frame structure, the frame header contains a timestamp and a sensor status code, and the frame body stores the binary data of the feature vector.

[0027] Example 2: refer to Figure 3 The implementation of the dynamic mapping layer is based on the time series matching algorithm of signal features. The pre-defined dynamic time warping path is used as the similarity anchor point set, which is composed of the time alignment points of historical reference signal feature combinations. Each anchor point contains the normalized values of the pressure spectrum feature vector and the motion trajectory feature vector, and the mapping relationship between anchor points is established through the time warping function. After the input of the real-time detection signal feature set, the dynamic time warping distance between it and each reference signal feature combination in the feature matching database is calculated. The distance calculation uses a sliding window comparison mechanism: the real-time pressure spectrum feature vector and the reference pressure spectrum feature vector are segmented according to the time sequence, and the Euclidean distance of the spectral energy distribution is calculated segment by segment; at the same time, the real-time motion trajectory feature vector and the reference motion trajectory feature vector are time-axis stretched and aligned, and the absolute value sum of the displacement mutation amplitude difference is calculated. The final distance value is the weighted sum of the spectral distance and the trajectory distance, and the weight coefficient is pre-set according to the pipeline type.

[0028] When the dynamic time warping distance is less than the preset similarity threshold, it is determined that the matching is successful. The system outputs an index number corresponding to the reference signal feature combination, which is directly associated with the optimal anti-shake parameter combination stored in the feature matching database. The index number uses a 16-bit encoding format, including a pipeline model classification code and a version identification code. When the real-time detection signal feature and all reference combinations have a distance exceeding the preset similarity threshold, a nearest neighbor projection mechanism is started. The mechanism traverses all reference combinations in the feature matching database, selects three candidate combinations with the smallest distance from the real-time feature, and generates a virtual reference signal feature combination based on the inverse proportion weighting of the distance values of the candidate combinations and assigns a temporary index number. The matching result (successful index or temporary index) is transmitted to the dual-domain interpolation calculation layer.

[0029] The operation of the dual-domain interpolation calculation layer is divided into two stages: feature domain interpolation and device domain interpolation. In the feature domain interpolation stage, the corresponding reference pressure spectrum feature vector is retrieved from the feature matching database according to the index number output by the dynamic mapping layer. If the index is a formal number of successful matching, a single reference vector is directly read; if it is a temporary index number, multiple reference vectors selected by the nearest neighbor projection mechanism are obtained. The multiple reference pressure spectrum feature vectors are weighted and fused: the fundamental component takes the weighted average value, and the weight is inversely proportional to the similarity distance; the harmonic energy distribution adopts vector addition operation, and each harmonic component is superimposed to generate a mixed frequency spectrum feature according to the weight coefficient. The mixed frequency spectrum feature retains the original vector dimension, including the fundamental frequency amplitude and the 2-5 harmonic energy values.

[0030] In the device domain interpolation stage, the reference motion trajectory feature vector is adjusted according to the installation angle parameter of the current blood transfusion pipeline. The installation angle parameter is obtained in real time by a dual-axis inclination sensor installed on the pipeline support, in the form of pitch angle and roll angle. The coordinate transformation process first establishes the conversion matrix between the pipeline local coordinate system and the global coordinate system. The displacement mutation amplitude data in the reference motion trajectory feature vector is converted from the standard laboratory coordinate system (Z-axis vertical upward) to the actual installation coordinate system: the displacement vector is rotated, and the rotation angle is consistent with the real-time inclination parameter. The time series correlation data is adjusted by resampling method to compensate for the time delay difference caused by the installation angle. The transformed motion trajectory feature vector includes the displacement mutation amplitude and the time series autocorrelation coefficient in the actual coordinate system.

[0031] The generation of the final anti-jitter control parameter adopts a dual-domain fusion strategy: the mixed frequency spectrum feature output by the feature domain and the coordinate transformation result output by the device domain are input into the parameter synthesis module. The fundamental frequency amplitude in the mixed frequency spectrum feature is directly mapped as a phase offset parameter, and the mapping relationship is a linear proportional function. The harmonic energy distribution array is processed by normalization and weighted summation to generate an amplitude suppression coefficient. The displacement mutation amplitude data output by the device domain is used to correct the amplitude suppression coefficient, and the correction amount is the deviation proportion of the amplitude and the preset threshold. The output anti-jitter control parameter includes a 32-bit floating-point phase offset parameter (unit: degree) and a 16-bit fixed-point amplitude suppression coefficient (resolution 0.1%). The parameter is packaged as a data packet and transmitted to the anti-phase jitter execution layer.

[0032] The maintenance of the dynamic time warping path adopts an incremental update strategy. After each successful match, the system records the time alignment path of the real-time detection signal feature and the matched reference combination. When the same path appears repeatedly for a preset number of times threshold, the path is added as a new anchor point to the similarity anchor point set. The anchor point set is stored in the cache memory, with an upper limit of 1000 groups, and the least recently used algorithm is used to manage the storage space. The candidate combination screening of the nearest neighbor projection mechanism uses spatial indexing acceleration, and an R-tree index structure is established in the feature matching database to realize fast nearest neighbor search.

[0033] The dual-domain interpolation calculation layer sets an exception handling process: when the tilt sensor data required for device domain interpolation is invalid, the last valid tilt parameter is enabled and the data exception flag is marked; when a zero-weight combination occurs in feature domain interpolation, switch to the default reference combination and trigger the parameter optimization layer to record the event. After the anti-jitter control parameter is generated, an effectiveness check is performed: the phase offset parameter range is limited to 0-360 degrees, and the amplitude suppression coefficient is limited to the 0.0-1.0 interval. When the check fails, the last period valid parameter is replaced, and an error code is sent to the exception decision layer.

[0034] The weighted fusion of feature domain interpolation adopts a double buffering mechanism: while the interpolation calculation of the current period is being performed, the reference vector data of the next period has already been preloaded into the buffer register. The device domain coordinate transformation uses a hardware acceleration module, which includes a dedicated floating-point operation unit to handle rotation matrix multiplication. The output interface of the parameter synthesis module uses a circular queue structure to ensure continuous output of the anti-jitter control parameter without interruption. The entire calculation process is completed within a fixed time window, and the calculation delay is controlled within 10 milliseconds.

[0035] Example 3: Referring to Figure 4The anti-jitter execution layer is implemented based on a combination of digital signal processing and electromechanical control technology. This layer receives anti-jitter control parameter data packets from the dual-domain interpolation calculation layer. These packets contain two core data points: phase offset parameters and amplitude suppression coefficients. The parameter parsing module first verifies data integrity: it checks the data packet header checksum to confirm that the phase offset parameter is within the valid range of 0-360 degrees and the amplitude suppression coefficient is within the range of 0.0-1.0. Invalid data triggers an error handling procedure, which replaces the parameters from the previous valid cycle.

[0036] The generation process of the inverted waveform command employs real-time signal reconstruction technology. Based on the phase offset parameter, the system extracts the fundamental frequency component and harmonic components from the original pressure waveform signal. The fundamental frequency component is locked using digital phase-locked loop tracking technology, and the harmonic components are separated using an adaptive filter. The reconstructed inverted waveform satisfies the following mathematical relationship: in: Represents a time variable. For the first The amplitude of the second harmonic ( (corresponding to the base frequency) For harmonic frequencies, The initial phase angle, This is the phase offset parameter. This formula ensures that the generated waveform maintains a precise 180° phase difference with the detected jitter signal. The waveform data is stored in a circular buffer in a 16-bit fixed-point format, with a sampling rate consistent with that of the pressure sensor.

[0037] The piezoelectric ceramic actuator employs a hierarchical drive architecture for control. The amplitude suppression coefficient is input to the gain adjustment module and converted into a drive voltage control quantity. The gain adjustment function is defined as follows: In the formula: This is the amplitude suppression coefficient. This is the calibration constant for the actuator sensitivity. This is the maximum safe drive voltage threshold. This function ensures a linear relationship between output gain and amplitude suppression coefficient. The drive circuit includes a high-voltage operational amplifier cascaded H-bridge topology, with an output voltage range of 0-200V and a ripple factor of less than 1%.

[0038] The phase compensation command integrates inverted waveform data with drive control parameters. The command data structure includes three fields: a waveform data pointer (pointing to a circular buffer address), a drive voltage value, and the waveform duration. The command encapsulation uses a time-division multiplexing protocol and is transmitted to the actuator node via a high-speed serial interface. The transmission cycle is synchronized with the signal detection window to ensure that the compensation action is completed before the next detection window.

[0039] The deployment of piezoelectric ceramic actuators adopts a distributed layout. Along the blood transfusion pipeline, a set of actuator units is installed every 20 centimeters, each set containing two orthogonally distributed piezoelectric ceramic sheets. The ceramic sheets are attached to the outer wall of the pipeline through epoxy adhesive, and the driving direction is consistent with the pipeline vibration direction. The built-in temperature sensor in the actuator unit monitors the working state, and automatically reduces the driving voltage when the temperature exceeds 85°C.

[0040] The execution process includes a dynamic calibration mechanism. Before each drive, the actuator feedback circuit measures the actual displacement of the ceramic sheet and compares it with the expected displacement of the command. When the deviation exceeds 5%, online calibration is started: adjust the driving voltage compensation coefficient, update the actuator sensitivity constant . Calibration data is recorded in non-volatile memory for subsequent driving precision optimization.

[0041] The safety protection mechanism sets multiple levels of restrictions: the driving voltage real-time monitoring module prevents overvoltage impact, the current limiting circuit sets a 150mA threshold, and the mechanical displacement sensor detects actuator overtravel motion. When any protection mechanism is triggered, the phase compensation command output is immediately interrupted, switched to passive damping mode and sends a fault code to the exception decision layer.

[0042] The generation process of the inverse phase waveform command includes frequency tracking fault tolerance processing. When the base frequency lock fails, the standby frequency generation mode is enabled: generate a reference waveform based on the historical base frequency data mean, while marking the frequency loss of lock state. In this state, the system reduces the amplitude suppression coefficient upper limit to 0.7 to avoid amplifying compensation errors and jitter.

[0043] The actuator node adopts distributed clock synchronization. Each node receives a unified time reference signal, and the compensation waveform output time error is less than 50 microseconds. The node-to-node communication adopts a daisy chain topology, with the master node sending global commands and the slave nodes returning status reports. The network delay compensation algorithm ensures that all actuators act synchronously.

[0044] After the phase compensation command is executed, the system enters a waiting state until the next detection period. This state maintains the actuator at standby voltage (10% of maximum driving voltage), reducing piezoelectric ceramic fatigue loss. The standby voltage value is dynamically adjusted according to the ambient temperature, and the standby voltage decreases by 1V for every 10°C increase in temperature.

[0045] The driving effect verification mechanism is embedded at the end of the execution layer. The built-in micro-accelerometer in the actuator collects the actual vibration waveform and performs cross-correlation analysis with the expected compensation waveform. When the correlation coefficient is less than 0.8, mark this compensation failure event and record the waveform difference data. When the cumulative failure events reach the threshold, trigger the actuator self-checking program.

[0046] The maintenance cycle of piezoceramic actuator is predicted by driving history data. The system accumulates driving time and driving energy integral value, and generates a maintenance reminder when it reaches 80% of the material life index. The driving energy integral calculation uses: Where: is the driving voltage, is the driving duration, is the total driving times. When this value exceeds the threshold, the maximum driving voltage is automatically limited to a safe value.

[0047] The interface between the execution layer and the closed-loop feedback layer sets up a data forwarding channel. After each phase compensation instruction is executed, the driving parameters and actual vibration data are forwarded to the feedback layer. The data includes time-stamped driving voltage values, waveform spectral characteristics, actuator temperature, and other parameters for residual jitter analysis.

[0048] Example 4: The implementation of the closed-loop feedback layer is realized through a multi-level monitoring loop. Within 50 milliseconds after the phase compensation instruction is output by the anti-phase jitter execution layer, the array pressure sensor reacquires the blood pipeline pressure waveform signal. The acquisition window length is set to 300 milliseconds, covering the complete compensation period. The residual jitter energy value is calculated using the piecewise integral method: the reacquired signal is divided by the fundamental frequency period, the root mean square value of the pressure fluctuation amplitude in each period is calculated, and the arithmetic average of all period results is taken. This value is quantified in millivolts, reflecting the strength of the residual vibration that has not been canceled.

[0049] The preset jitter threshold parameter is set by the initial control instruction generation layer and stored in the non-volatile memory. The real-time comparison process is realized in a dedicated comparator circuit: the residual jitter energy value is input to the positive terminal of the comparator, and the threshold parameter is input to the negative terminal. When the residual value continuously exceeds the threshold for three consecutive sampling periods (totaling 900 milliseconds), the comparator outputs a high level to trigger an interrupt signal. The interrupt service program generates a feature matching database update request, which contains the following data structure: Timestamp (accurate to milliseconds); Residual energy value (floating point type); Corresponding pipeline number (ASCII string); Current reference signal combination index (16-bit encoding).

[0050] The update request is passed to the parameter optimization layer through the message queue, and the queue capacity supports caching 100 requests.

[0051] The parameter optimization layer receives the historical residual jitter data from the closed-loop feedback layer. The residual jitter energy time series is stored in a ring buffer with a depth of 120 groups of data, covering the last 10 minutes of records. Each group of data contains a timestamp, residual energy value, ambient temperature, and flow rate parameters. The gradient boosting decision tree model is constructed using a sliding window training mechanism: the last 60 groups of data are used as the training set, the fundamental frequency amplitude, second harmonic ratio, and fifth harmonic slope in the reference pressure spectrum feature vector are used as feature variables, and the residual energy value is used as the target variable. The model contains 50 decision trees, with a maximum depth limit of 6 layers. The training process uses the gradient residual optimization algorithm, and each additional 10 groups of data triggers incremental model updates.

[0052] The association rule analysis focuses on the influence weight of the feature variables on the target variable. When the correlation coefficient between a specific reference pressure spectrum feature vector (such as fundamental frequency amplitude > 2.5 mV and second harmonic ratio < 15%) and high-frequency residual jitter (> threshold 150%) exceeds 0.85, it is marked as a strong association feature combination. At this time, the feature matching database optimization command is generated, which contains the following instruction set: Feature combination identification code (based on feature vector hash value); Optimization type label (feature replacement / parameter adjustment); Association strength coefficient (0.0-1.0).

[0053] The optimization command needs to pass a consistency check before triggering the full-cycle update mechanism: check the frequency of the same feature combination in the historical optimization record to avoid repeated updates. After passing the check, the command is written to the system instruction register. See Table 1.

[0054] Table 1: Residual jitter energy monitoring record table.

[0055] Time stamp Residual energy value (mV) Threshold parameter (mV) Out-of-limit flag Ambient temperature (°C) Flow rate (mL / min) 2023-08-25 12.8 10.0 1 24.5 85 2023-08-25 13.2 10.0 1 24.6 86 2023-08-25 12.9 10.0 1 24.7 85 2023-08-25 9.8 10.0 0 24.6 84 2023-08-25 10.2 10.0 0 24.5 85 The feature importance evaluation of the gradient boosting decision tree model uses the gain contribution algorithm. After each model training, the feature variable importance ranking table is output and stored in the optimization log file. When the importance score of the fundamental frequency amplitude feature exceeds 0.65 for three consecutive times, the emergency optimization process of the reference signal feature combination is automatically triggered. This process bypasses the regular association analysis and directly generates the feature matching database optimization command.

[0056] The historical data management system sets an automatic archiving strategy. Every 24 hours, the ring buffer data is transferred to the historical database, and the optimization logs older than 30 days are deleted. The database index is established based on dual B-tree indexes of time range and feature combination, supporting fast retrieval of residual jitter patterns under specific working conditions.

[0057] Conflict detection before optimization command execution: Check if there is an unfinished optimization task for the same pipeline number. Enter the queuing state when there is a conflict, and the task scheduler handles it according to priority. High-priority tasks (such as emergency optimization processes) can interrupt low-priority tasks to ensure that critical updates are executed in a timely manner.

[0058] The interface between the parameter optimization layer and the feature matching database uses atomic operation protocol. The update process starts a transaction lock mechanism to prevent read-write conflicts. A version change record is generated after each optimization command execution, containing metadata such as old feature combination hash value, new feature combination storage address, and change time. The version record is used to support rollback function, when the new benchmark combination causes residual jitter to increase by more than 20%, automatically restore to the last valid version.

[0059] Example 5: Implementation of the abnormal decision layer is based on multi-level state machine judgment logic. The system continuously monitors the residual jitter energy value sequence output by the closed-loop feedback layer, with a sampling interval of 300 milliseconds. The condition for determining the state of uncontrolled jitter is set as follows: the residual jitter energy value in three consecutive detection periods exceeds 150% of the preset jitter threshold parameter. The detection period counter uses a sliding window mechanism, with a window length of three periods. When all residual values in the window meet the over-standard condition, the state flag register is set. The preset percentage parameter is stored in programmable memory, supporting dynamic adjustment of threshold proportion according to pipeline type.

[0060] The generation of pipeline safety risk level assessment report uses a multi-dimensional analysis model. The report data structure includes: current residual energy value, over-standard duration, historical over-standard frequency, environmental temperature, and liquid viscosity parameters. The risk assessment matrix maps the above parameters to four risk levels: Level 1 (low risk): residual value exceeds the standard but the trend is stable Level 2 (medium risk): residual value continues to rise Level 3 (high risk): residual value fluctuates sharply Level 4 (critical): residual value is close to the pipeline rupture threshold The report is automatically attached with a timestamp and pipeline location code after generation, and uploaded to the central monitoring system through the medical device communication protocol.

[0061] The implementation of the sound-light alarm protocol uses a hierarchical response strategy. The primary alarm activates a yellow warning light and an 800Hz intermittent buzzer, and if there is no response for 10 seconds, it is upgraded to secondary alarm: a red rotating warning light with a 1200Hz continuous buzzer. The alarm signal is transmitted synchronously through the ward broadcast system. The emergency shutdown instruction contains an encrypted shutdown code and pipeline number, and is sent to the peristaltic pump control system in ModbusRTU format. The shutdown process uses a ramp-down mode, with the flow rate linearly decreasing to zero within 5 seconds to avoid liquid impact.

[0062] The trigger of the full-cycle update mechanism relies on the feature matching database optimization command generated by the parameter optimization layer. After verifying the digital signature, the command parsing module extracts the real-time detection signal feature set corresponding to the latest residual jitter energy value in the time period. The time period range is locked to the three detection cycles (900 milliseconds) before the determination of the uncontrollable jitter state. The data extraction process calls the ring buffer backtracking function, and retrieves the pressure spectrum feature vector and motion trajectory feature vector according to the timestamp.

[0063] The writing operation of the new reference signal feature combination follows a strict data verification process. The verification content includes: whether the fundamental frequency of the pressure spectrum feature vector is within the physiological allowable range of 1-10Hz, and whether the displacement mutation amplitude of the motion trajectory feature vector exceeds the material elastic limit. After verification, the system assigns a unique index number to the new combination, and the number structure contains the first 8 bits of the timestamp hash value. The writing operation is executed in a database transaction: first, copy the original reference combination data structure, replace it with the real-time detection signal feature set, and finally append the initial estimated value of the optimal anti-jitter parameter combination. The initial parameters are generated by the historical mean algorithm, marked as to-be-optimized state.

[0064] The reset process of the similarity anchor point set updates the matching reference of the dynamic mapping layer. The system calculates the minimum Euclidean distance between the new reference signal feature combination and the existing anchor points, and if the distance exceeds the preset threshold, the combination is added to the anchor point set. The anchor point set is stored in a dynamic array, and the addition of new anchor points triggers a reordering operation: a binary search tree index is established according to the pipe model to optimize the nearest neighbor search efficiency. After resetting, the matching range of the dynamic mapping layer is expanded to the feature space containing the new combination.

[0065] The false alarm correction mechanism is set in the abnormal decision layer. After the emergency stop command is issued, the system continuously monitors the pipe pressure change. If the pressure returns to below the safety threshold within 30 seconds, a false alarm analysis report is automatically generated and the alarm state is cleared. The report contains the sensor raw data and environmental parameter snapshot at the time of triggering, which is used for subsequent algorithm optimization.

[0066] The full-cycle update mechanism includes a version rollback function. A database snapshot is created before each update and stored in a separate storage partition. If the new reference combination triggers an increase in residual jitter anomaly (more than 20% of the historical mean) within the next 24 hours, the system automatically rolls back to the previous version and marks the combination as invalid data. The rollback operation updates the similarity anchor point set synchronously, removing anchor points related to invalid combinations.

[0067] The execution confirmation of the emergency stop command uses double feedback: the peristaltic pump control system returns a stop confirmation code, and the pressure sensor verifies the flow rate zero state. If no confirmation code is received or the flow rate is not zero, the system reissues the command every 5 seconds until the maximum number of attempts is reached. If the maximum number of attempts is exceeded, the mechanical emergency braking protocol is triggered, activating the pipe physical clamping device.

[0068] Risk level assessment reports are integrated with the hospital information system. Level 4 critical reports automatically trigger a nursing station priority handling queue, while a red alert entry is generated in the electronic medical record system. Report data retention period is 90 days, supporting historical risk event retrieval by time range.

[0069] The deactivation of the audible and visual alarm protocol requires a double confirmation: after the medical staff on-site key confirmation, the system verifies that the residual dithering energy value has been below the safety threshold for 60 seconds. The deactivation signal is sent synchronously to the central monitoring system to update the status board.

[0070] It should be noted that, in the present document, relational terms are used solely to delineate the scope of the inventive subject matter and the terms are not intended to exclude the equivalence of what is described and / or claimed, unless specifically indicated. Rather, the scope of the inventive subject matter is defined by the appended claims.

[0071] While the embodiments of the application have been illustrated and described, it will be understood by those skilled in the art that various changes, modifications, substitutions, and alterations can be made therein without departing from the spirit and scope of the application, which is defined by the following claims and their equivalents.

Claims

1. A smart sensor-based transfusion line pulsation detection method, characterized in that, The application relates to a blood transfusion pipeline anti-vibration control method based on multi-modal signal detection and dynamic mapping. The initial control instruction generation layer receives operation instructions, identifies pipeline numbers and preset vibration threshold parameters through a natural language analysis unit, and the like. The multi-modal signal detection layer is used for real-time acquisition of pressure waveform signals and three-dimensional motion trajectory signals of the blood transfusion pipeline. The feature matching database is used for storage of reference signal feature combinations corresponding to different pipeline models. The dynamic mapping layer is used for similarity matching of real-time detection signal features and the reference signal feature combinations. The dual-domain interpolation calculation layer is used for generation of anti-vibration control parameters according to the similarity matching results. The anti-phase vibration elimination execution layer is used for output of phase compensation instructions according to the anti-vibration control parameters. The parameter optimization layer is used for updating of the feature matching database according to historical vibration residual data. The abnormality decision layer is used for judgment of an un-suppressible vibration state and triggering of an alarm protocol.

2. The method of claim 1, wherein the method is based on a smart sensor. The multi-modal signal detection layer comprises the following steps: frequency spectrum distribution features of pressure waveform signals are acquired through an array type pressure sensor; time domain mutation features of three-dimensional motion trajectory signals are captured synchronously through an inertial measurement unit; the frequency spectrum distribution features and the time domain mutation features form a real-time detection signal feature set; and the real-time detection signal feature set is transmitted to the dynamic mapping layer for similarity matching.

3. The method of claim 1, wherein the method comprises: The feature matching database comprises the following steps: a three-dimensional parameter architecture is established for each pipeline number; a first dimension is used for storage of reference pressure frequency spectrum feature vectors; a second dimension is used for storage of reference motion trajectory feature vectors; and a third dimension is used for storage of optimal anti-vibration parameter combinations; the reference pressure frequency spectrum feature vectors comprise fundamental frequency and harmonic energy distribution data; and the reference motion trajectory feature vectors comprise displacement mutation amplitude and time sequence correlation data.

4. The method of claim 1, wherein the method is based on a smart sensor. The dynamic mapping layer comprises the following steps: a dynamic time warping path is predefined as a similarity anchor point set; dynamic time warping distances of real-time detection signal features and reference signal feature combinations are calculated; when the dynamic time warping distance is smaller than a preset similarity threshold value, an index number of a matched reference signal feature combination is output; and when the preset similarity threshold value is exceeded, a nearest neighbor projection mechanism is started to select a reference signal feature combination with the smallest distance.

5. The smart sensor-based transfusion line pulsation detection method of claim 4, wherein, The dual-domain interpolation calculation layer comprises the following steps: in a feature domain interpolation stage, a plurality of reference pressure frequency spectrum feature vectors are weighted and fused to generate a hybrid frequency spectrum feature based on the index number of the matched reference signal feature combination; in a device domain interpolation stage, a reference motion trajectory feature vector is coordinate-transformed according to an installation inclination angle parameter of the current blood transfusion pipeline; and finally, the hybrid frequency spectrum feature and the coordinate transformation result are fused to generate anti-vibration control parameters.

6. The method of claim 1, wherein the method is based on a smart sensor. The anti-phase vibration elimination execution layer comprises the following steps: anti-vibration control parameters generated by the dual-domain interpolation calculation layer are received; an anti-phase waveform instruction is generated according to a phase offset parameter in the anti-vibration control parameters; an output gain of a piezoelectric ceramic actuator is adjusted according to an amplitude suppression coefficient; and the phase offset parameter and the amplitude suppression coefficient jointly constitute a phase compensation instruction.

7. The method of claim 1, wherein the method is based on a smart sensor. The application further comprises the following steps: The closed-loop feedback layer is used for real-time monitoring of residual pipeline vibration after compensation. The implementation mode of the closed-loop feedback layer comprises: after the phase compensation instruction output by the inverse de-bouncing execution layer, re-acquiring the pressure waveform signal to calculate the residual jitter energy value; The residual jitter energy value is compared with the preset jitter threshold parameter in real time; when the residual jitter energy value is continuously higher than the preset jitter threshold parameter, a feature matching database update request is generated.

8. The smart sensor-based transfusion line pulsation detection method of claim 7, wherein, Also includes: The parameter optimization layer updates the feature matching database according to the historical jitter residual data; The update mechanism of the parameter optimization layer comprises: receiving the residual jitter energy value time sequence generated by the closed-loop feedback layer; constructing a gradient boosting decision tree model to analyze the association rules of the residual jitter energy value and the reference signal feature combination; when a specific reference pressure spectrum feature vector is detected to be related to high-frequency residual jitter, a feature matching database optimization command is generated; the feature matching database optimization command triggers the full-cycle update mechanism.

9. The method of claim 1, wherein the method is based on a smart sensor. The judgment logic of the abnormal decision layer comprises: when the residual jitter energy value exceeds the preset jitter threshold parameter by a preset percentage in three consecutive detection cycles, it is determined to be an uncontrolled jitter state; a pipeline safety risk level evaluation report is generated; a sound-light alarm protocol is triggered and an emergency shutdown instruction is sent to the peristaltic pump control system.

10. The method of claim 8, wherein the method further comprises: The execution steps of the full-cycle update mechanism comprise: receiving the feature matching database optimization command generated by the parameter optimization layer; extracting the real-time detection signal feature set corresponding to the latest residual jitter energy value period; writing the real-time detection signal feature set as a new reference signal feature combination into the feature matching database; resetting the similarity anchor point set coverage range of the dynamic mapping layer.

Images